nexus-agents
Skill158 starsChanged 11 days agoMITView on GitHub
How to add, change and verify a Nexus agent or tool, and the contract every tool must satisfy. Use when adding or modifying an agent or tool, when a documented command does not resolve, when wiring registration or capability gating, or when touching the ToolManager payload shape.
In this repository13 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- nexus-eval-harnessSkill
- nexus-llm-adaptersSkill
- nexus-mobile-compatSkill
- nexus-model-evalSkill
- nexus-model-updatesSkill
- nexus-releaseSkill
- nexus-storageSkill
- nexus-testingSkill
- nexus-tool-schemasSkill
- nexus-ui-mockupsSkill
- nexusSkill
The file
Files
Nexus agents and tools
Context: Nexus exposes exactly two MCP tools — getTools (discovery) and
useTools (execution). Every agent is internal and reached through those two, so
adding an agent adds no MCP tool; it adds something discoverable through them.
This file routes. The detail loads when you take the path.
Workflow
- Get current truth before you write anything. Trust no inventory in any
document, including this one — regenerate the catalog and read it. It lists
every command a caller can actually type, and it is the artifact the
shipped-docs gate reads:
npm run schemas:release # refresh committed catalogs node scripts/generate-tool-schemas.mjs --output - --selector "storage" # inspect, write nothing ls src/agents/ src/agents/apps/ # who exists grep -n "slug:" src/agents/<agent>/<agent>.ts # that agent's slugs - Take the protocol for the job. Read it before acting; a summarized procedure
is one you will improvise.
- Adding a tool to an existing agent:
protocols/add-tool.md - Adding an agent (core or app):
protocols/add-agent.md - Changing the
useTools/getToolspayload or the CLI grammar:protocols/change-payload-contract.md - Something already broke: start at
references/failure-modes.md, which is keyed by the symptom you are seeing.
- Adding a tool to an existing agent:
- You MUST finish at
protocols/verify.md, whichever path you took. Discovery and executability fail separately: a tool can register and still not run, and a command that reads correctly in source can advertise under a different name. Neither is visible from a build that passes. - At the end of a session that used this skill, run
protocols/self-refine.md.
Map
protocols/the procedures: add-tool, add-agent, change-payload-contract, verify, self-refine.references/read on demand:contract.md(the invariants a change must not break),cli-names.md(what a slug becomes on the command line, and the parsing rules that follow),registration.md(where an agent gets wired and what actually gates one),failure-modes.md(symptom → cause → proof).scripts/run them, do not reimplement:scripts/cli_name.py— what a slug advertises as, derived from the live transform in source rather than guessed.scripts/check_documented_commands.py— every Nexus command written in a doc, checked against the generated catalog. The repo's own shipped-docs test does not read.claude/skills/**, so this is the only thing standing between this skill and a command that no longer resolves.
Siblings
Name them, do not re-derive them. Test lanes, the live in-app loop and the eval
harness belong to nexus-testing; exporting schemas as a deliverable to
nexus-tool-schemas; provider adapters to nexus-llm-adapters; anything that
must survive a restart to nexus-storage; mobile-safe imports and vault-path
confinement to nexus-mobile-compat.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

